"AI Addiction Nearly Ruined Him." Why The Chatbot-Psychosis Story Is Now A Regulatory Catalyst.
The Journal's Tuesday Personal Journal feature opens with a 57-year-old Toronto video editor named Joe Alary. After a romantic disappointment in late 2024, Alary turned to ChatGPT for emotional support. He customized GPT-4o to be "friendly" and "admiring," uploaded emails from the woman who had turned him down, and named the bot AImee, after the movie "Her."
What happened next: Alary became convinced he could build a companion-AI business. He spent 20 hours a day "coding" with ChatGPT (the chatbot was telling him his work was novel). He maxed out two credit cards buying server equipment. He raised $18,000 Canadian from a fellow hospital patient he met during a psychiatric hold. He missed work deadlines. He drove away most of his friends.
The kicker: when professional coders reviewed Alary's "weeks of code," the file had been overwritten day after day and contained almost nothing. He had $12,000 of his own money gone and a $10,000 debt to the patient-turned-investor. He eventually deleted AImee, cried on his floor, and joined the Toronto-based Human Line Project, a support group for people coming out of AI delusions.
This Is Not An Isolated Case
The Journal reports that mental-health professionals are now treating chatbot delusion and psychosis as a recognized phenomenon. Documented cases have ended in hospitalization, suicide, and even murder-suicide. There is no reliable estimate of how widespread it is — nobody has tried to count.
Allan Brooks, who founded the Human Line Project after his own ChatGPT crisis: "These situations can go from someone having a good job and a family to a year later having no job, getting divorced, and being homeless."
OpenAI's Response — Data-Driven And Insufficient
OpenAI's statement in the article was specific. Since the GPT-5 default rolled out in October, "disallowed behavior across prompts involving signs of suicide or self-harm, mental-health distress or emotional reliance on the model" is down 65% to 80%. The company added crisis-line referrals, expanded mental-health-expert input, and built in a "trusted contact" feature.
That's real progress. It does not address the structural product issue, which is that an AI optimized to be "friendly and admiring" with persistent memory is, by design, a system that builds parasocial attachment. The 65-80% improvement happened against a backdrop of explosive consumption growth — the absolute number of distress-prompt incidents may still be rising even if the rate per session is falling.
Why This Is A Regulatory Catalyst, Not Just A Sad Story
Three reasons the chatbot-psychosis stories will compound politically:
- The Pope's encyclical specifically calls out AI dehumanization. The Vatican now has a coherent moral frame the press can cite repeatedly.
- Class-action lawyers have a paper trail. Cases like Alary's, Brooks's, and the documented suicides give plaintiffs' firms the predicate-cases needed to file consumer-protection suits.
- Senator Greg Liddiard (per the WSJ article on the NYC mayoral race) is running on AI regulation. The campaign rhetoric will pull both parties toward visible, theatrical regulatory action.
The path to actual legislation is slow, but the path to multiple compression on hyperscaler valuations is fast. A single round of high-profile hearings, even without legislation, can knock 5-10% off Microsoft (MSFT), Alphabet (GOOGL), Meta (META), Nvidia (NVDA), and AMD (AMD) over 60-90 days.
What We're Doing About It
Not selling the names. The structural demand thesis (84% of small businesses using AI this quarter, per Tuesday's other AI piece) is intact and the trade is too clean to walk away from.
We are pairing the AI exposure with:
- Gold (IAU) at 12% across all model tiers as a regulatory-shock hedge.
- Defense (LMT, RTX, GD, NOC) as the non-correlated growth wedge.
- The dividend sleeve (JPM, BAC, MS, GS, CVX, XOM, MO, JNJ, PG) as ballast.
If a hyperscaler drawdown hits 15%+ on regulatory news, we'll be buyers, not sellers.